From Malay Mail · · 2m
100 petabytes of Nasa data, one minute per forecast
Swiss researchers parked all of Nasa's climate data next to a supercomputer to catch disasters before they happen.
In brief
Swiss researchers parked all of Nasa's climate data next to a supercomputer to catch disasters before they happen. Forecasts that once took hours now run in about a minute and cost little to repeat, so warnings can be rechecked every few minutes instead of once a cycle. Originally reported by Malay Mail.
A forecasting revolution
Researchers at ETH Zurich are training AI on huge stores of Nasa climate data, chasing potentially lifesaving forecasts at speeds the old models cannot touch.
Patterns humans miss
Speed is only half of it. AI can pull signals out of satellite data nobody could see before, which could flag events like Nepal's flood that left thousands dead or missing.
The two things it takes
Training these models demands mountains of high quality climate and Earth observation data, plus serious computing power. Almost nobody has both in one place.
100 petabytes, next door
ETH copied roughly 100 petabytes of public Nasa data onto servers sitting beside Alps, one of the world's most powerful supercomputers, in Lugano.
“This is a huge scientific opportunity”
A year just to copy it
Six billion files took about a year to transfer. Climate physicist Reto Knutti puts that at roughly 20 million feature length films, or a million typical computers' worth of storage.
“Data is essentially everything”
Seconds, not days
Those servers hum a few metres from the machine for a reason. Waiting on transfers would stall the work before any science starts.
“It matters whether you can move the data within a few seconds or whether you have to wait days for the data to come”
Statistics beat equations
The traditional method simulates ocean and atmosphere physics with mathematical equations, and takes hours. A pattern recognition model runs a multi day global forecast in about a minute.
Cheap enough to rerun
Training costs a lot, running costs almost nothing. That lets scientists repeat a forecast every few minutes to check whether a dangerous pattern is forming.
“That allows us to do early warning systems to save lives.”
Blatten was visible early
The Swiss village was wiped out by a glacier collapse in May last year. Close monitoring let authorities evacuate a week before it happened, avoiding mass casualties.
“that was visible in satellite data more than a year before it actually happened”
The signs in Nepal
Nature reported that satellite imagery showed warning signs before the August 26 glacial collapse on the Nepal China border. Caught in time, the area could have been flagged for closer watching.
Watching, not waiting
Knutti's pitch is systematic observation instead of luck: monitoring systems that scan the data continuously for hazard signals, and could save hundreds or thousands of lives.
The main idea
Forecasts that once took hours now run in about a minute and cost little to repeat, so warnings can be rechecked every few minutes instead of once a cycle.
